Leading AI Economist: AI Price War Is Getting Worse | Fortune AI Weekly

AI spend essentially flatlined in August and September. We saw virtually no growth in AI spend. What's really going on with AI spending? Our guest this week gives us an answer you might not expect. That was entirely driven by slowdowns among the top 1% of firms. Also, Democrats swept both the House and the Senate in the midterm elections. We'll discuss what this means for the AI boom. I think it's going to be a lot harder to build new data centers, and that obviously could really slow down the development of this technology. It could potentially really hurt. Some of these companies that are involved in the AI boom, it's not clear that OpenAI or Anthropic are going to get the compute capacity that they really need. I'm Jeremy Kahn, and we're going to talk about all of this on today's Fortune AI Weekly. Fortune AI Weekly is brought to you by ServiceNow. ServiceNow's AI specialists do the work, start to finish, so people can do the parts of the job they're best at. To learn more, visit ServiceNow.com. One of the questions we're always talking about is what's actually happening with AI spending at big companies? You know, who's winning the market share race? Is it OpenAI? Is it Anthropic? What's actually happening? Are people really using open source models? Well, to help us answer all those questions, we've got Ara Karazian, who is the lead economist at Ramp, joining us on the podcast today. Very excited to have him. Ara, for those that don't know you, do you want to give us a quick intro to you and Ramp and your work just to start off with? Sure. So I'm an economist at Ramp. You can think about the research that we do in terms of products that we cover in our data set. So Ramp is an AI finance platform, 85,000 plus businesses, 275 billion plus a year in annual spend. And so we see that whole universe of business spend in our data set. That's what I get to work with from ACH, corporate card invoicing. So you can imagine that we have this great, highly granular and highly responsive data set that allows us to see where businesses are spending more, where they are pulling back. And over the last two years, we have been tracking AI spend. It is the fastest growing spend category we have ever observed. And so over the last year, we've tracked both the adoption of AI and its intensity as it proliferates throughout the economy and trying to understand what makes some firms work better with it, what makes some firms hesitant to adopt it at all, and what does that all mean for the market and how much we should invest in AI as a larger economy. In terms of the data you have about AI adoption, what are we seeing in terms of how this technology is diffusing across the economy? And then what are we seeing in terms of whether firms are actually starting to see a return on investment from AI? Well, I do think this is one of the things that makes it very hard as an economist to measure AI's impact, right? In that if you look at most productivity statistics that are aggregated across the economy, you don't really see much of anything. Now, I do think, however, if you were to look at very specific pockets of the economy, and if you had a highly granular data set, you actually do see some impact. Firms that use AI are already fast-growing firms. Because they tend to be more techie, but they tend to accelerate upon adoption of AI, they tend to grow faster in terms of both their revenues, they tend to hire more, that's what the research is starting to show. And much of that is because of specific factors to how AI has proliferated throughout the economy. 1% of businesses drive 80% of the revenues to OpenAI and Anthropic. So it's extremely concentrated. The median firm spends only $12 to $15 per employee per month on AI, so that's about a chat subscription per employee. The top 10% firm spends anywhere from like $600 to $1,000 per employee per month. The top 1% firm, however, spends $10,000 per employee per month. So if you are analyzing this market, you need to have a highly granular data set that allows you to subset different types of businesses from each other based on their intensity of adoption. It's not really enough to just say, hey, we're adopting AI and therefore, you know, here's how we're growing because of it. It really does start to matter how the firm is starting to use it. And so far that's meant, you know, are they using it in a productivity enhancing way? Are they using the coding agents versus just the simple chat subscriptions? Are they spreading their spend across multiple models? Are they being thoughtful about how they're getting ROI in terms of, you know, routing some tasks over to cheaper models and using the frontier models only when necessary? Those are a lot of the questions that we try to get at with our work. Are you seeing with the firms that are spending more, the ones that are, you know, 600 or more dollars per employee, are they also the firms that are growing revenue the fastest? So we wrote a paper about this over the summer where we, you know, everybody wants to write the paper that says how AI is affecting the job market. But in order to do that, you do kind of need some data set that says, hey, these firms adopted AI, these firms didn't. What happened pre-post? That's surprisingly hard to get. It's also very hard to define. But we have the spend data set that tells you who's adopting AI and how. We combine that with a workforce data set from Revelio Labs. And we found exactly that, that firms that adopted AI, they're already faster growing from the get-go. I think that is pretty straightforward and makes sense to people. However, their adoption of AI leads to this acceleration in their hiring growth. They tend to grow about 10% over the two years following adoption. However, that growth is entirely concentrated amongst high intensity firms. So we define that as like firms that spend in the top third of AI per employee per month. They were spending at that time about $30 per employee per month at the beginning. It rises thereafter, right? But it's not like thousands of dollars is what I mean. Whereas the sort of Median AI spenders saw essentially no gain. And this is really kind of the sort of difficult fact of figuring out how to distribute the gains of AI throughout our economy is that today, those high intensity firms are also pretty much tech companies or they're technical services companies or they're involved in finance or something, right? They tend to be in these sectors that are uniquely able to take advantage of the most commercially advanced productivity enhancing version of AI today, which is coding agents. They have clearly shown some gain to those sectors, but there remains an open question whether or not that productivity gain will be able to proliferate through the rest of the economy into non-technical sectors. And so as we devise policy solutions, one of my great frustrations has been there seems to be a lot of focus around what's going to happen to tech companies or software engineers or certain categories of white-collar work. Based on the evidence so far, it actually looks like many of those categories are going to be growing just fine. They may even accelerate in terms of headcount and revenue growth. And where we should be worried about are sectors like manufacturing or construction, where there are actual reasonable, maybe a couple of years down the line, long-term impacts of AI that will accelerate the manufacturing automation we've already observed over the last century. But that I think our discourse is relatively underrating. I saw you recently wrote something saying that AI adoption might be starting to plateau a bit, or you're starting to see some evidence that perhaps, as you said, there was some weakness in kind of the foundations of the AI trade, at least in terms of the financial markets. Can you explain a little bit about what you meant by that? You know, I have no projections myself around where OpenAI or Anthropic IPOs should land. And if you take an economist's perspective on AI. You're really just looking at long-term productivity gains. You don't really care too much about what's going to happen in the stock market or to the companies that develop it in terms of their own profits. And still, though, I talk to a lot of investors and I talk to other people in the tech world about their projections around AI spend. And as I was entering these conversations, I found that they were making these assumptions that I didn't think were particularly grounded in the data. And so I wanted to... provide in this series of posts some data that I thought would either confirm or correct those assumptions if need be. And so that is one of them. One of them being that if you have this notion that AI spend is going to rise faster and faster and faster, even amongst the firms that are spending a lot on it already, not to say that won't happen, but if you were to look at the data today, that spend is already starting to slow down. AI spend essentially flatlined In August and September, we saw virtually no growth in AI spend. That was entirely driven by slowdowns among the top 1% of firms. So there's still growth at the median in the top 10%. But remember, those firms are spending less than 20% of the revenue that actually goes to OpenAI and Anthropic. So they're not yet a meaningful part of the market. And then if you look at what's happening at the top 1%, it's not a demand issue. It's not like they're not seeing value from AI. If anything, it looks like they are. They're using AI more and more per month, but AI is itself getting really cheap. The price per million tokens, so the literal pricing that OpenAI and Anthropic offer on their models, has essentially halved in the last six months. And that's because of a combination of price cuts and aggressive competition between OpenAI and Anthropic at that frontier level. So trying to make better models for a lower price to gain market share ahead of their own IPOs. And it's also because of increasingly performance and cheaper models at the standard and light levels. So if you don't need the best model for something, you can use a sort of cheaper model. And those models are getting cheaper. And so people will say like, oh, well, you know, they're just trying to get market share. It's like Uber and Lyft, they're just cutting prices now and then they're going to jack them up and then they'll be fine. And it's like, you can't assume that because if anything, the market's getting even more competitive. Muse, you know, is sort of Meta's latest consumer product for AI, but they're now getting into the enterprise market and offering models to enterprise that are also pretty cheap and highly performant. NVIDIA, too, is getting into the race. Grok, which is SpaceX AI, Elon Musk's thing, is also starting to offer pretty highly performant enterprise models at a low cost. And so I only see these dynamics getting cheaper, I mean, getting fiercer. All to say, We may end up in this market in which businesses use a lot of AI. They use it more cheaply in the form of cheap frontier models and also cheaper standard and light models. That's great for the businesses that use it, but it poses new questions for OpenAI and Anthropic that they didn't necessarily have to think about before around. How are they going to keep their market share and how are they going to continue to grow and invest in themselves? That's super interesting. I feel like we've spoken about the price war stuff before. Would you say that we're already kind of in the midst of a price war between these leading AI companies at the moment? A price war that's only getting stronger and stronger. If you look too at the price chart between OpenAI and Anthropic, you'll see that maybe about six months ago, Anthropic was both known as being a little bit more expensive. They had pretty high sticker prices on their models. And they had a little bit more, you could argue, stickiness in that engineers and high-intensity spenders really liked using Anthropic models. And so Anthropic was able to command a high price for that. That all changed this summer. Anthropic launched Fable, which was kind of a botched launch. It was the model that was so, so strong that the U.S. government decided to briefly ban it. It was also a really expensive model. And it was essentially around that time, maybe like a week or two after, that OpenAI came out with its own really highly performant frontier model. This was the sort of GPT-4 series of models. And what we ended up seeing in our dataset was Fable never really took off. It was a little too expensive at that respective price point. 4 was a worthwhile competitor. And a lot of the market shares start to shift over toward OpenAI as a result. Now then, Anthropic's response has been to aggressively price cut. Both in the form of sticker price cuts and also now enterprise discounts that they're offering on contracts to customers. And now if you look at the price chart, OpenAI and Anthropic are pretty much in line with each other. So if Anthropic did have an edge before in terms of stickiness of its businesses that wanted to use Anthropic models, It's hard to see evidence of that today. I just wonder what this does to the wider economics of the entire AI boom, though, because if they start cutting prices in this way, you know, we've seen compute costs, especially for like agentic work, is going up. So if the industry starts engaging in these really aggressive price wars, how do they balance that out against these rising costs? Well, this is where there is some disagreement amongst economists and analysts in the market, because I do think you could argue both sides. You could say. Hey, the top buyers of Compute today are OpenAI and Anthropic, and they are going to see some margin compression because competition is going to reduce their profits. Therefore, they will be less willing to pay for Compute. Okay, I think that that would probably put some downward pressure on Compute prices. At the same time today, there are perhaps enough sufficient other buyers of Compute that will be able to take their place. Not to say that OpenAI or Anthropic dropping out of the market wouldn't or sort of reducing their demand on that market wouldn't have an impact, but it may not have as large of an impact as we expect because there are other companies that are otherwise well capitalized, also want to grow their market share and aren't under that sort of same margin compression issue. Meta is going to be another buyer. SpaceX AI is already a big buyer. So you can end up seeing a lot of demand from there. As long as business demand continues to increase for AI models, and it looks like today, there is still more demand. There's going to be more demand growth from that top 10% and median profile of firms. Is it going to be that hyper growth phase that we saw in the top 1% over the last year? I think not unless we see some new commercial advancement in AI, something like the coding agent revolution that we saw over the last two years. Maybe we'll see that with consumer AI products. But that remains to be seen. On the enterprise use, are we starting to see even some of these sort of power users cutting back? We've seen reports just in the last week, like Goldman Sachs saying that they were pulling back access to, I think, Claude CoWork for most of their workforce, not actually their coders and engineers. They're still allowed to use it, but they had rolled CoWork out kind of widely across the firm, and now they've pulled that back. And I think there was just a report kind of similarly from Apple saying that they They were going to continue to allow certain engineering teams to use, you know, quite a lot of tokens. But basically, they're pulling everybody back from using, again, some of the more frontier models and some of these co-work products. Maybe about a year ago or less than a year ago, it was quite fashionable to, as a company, to be out there and saying and proudly state that you're spending a lot on AI and your employees are driving massive AI bills. And... That you're becoming an AI-pilled company. I don't think that's as fashionable today. Not that it's not fashionable to use AI today. I think most companies still want to use it in intelligent, smart ways that generate return on investment. But that return on investment part has become comparatively much more important. And if anything, if you are communicating to your investors how you're using AI, they are now really impressed to see that you are using AI, but you're using it cheaply and more efficiently with every passing month. And so if you saw six months ago, you know, it wasn't uncommon to see businesses proudly proclaim and post their AI bills and the growth trajectory in them. And now we're kind of seeing the opposite. The Notion CTO just the other day posted their AI bill, sort of they posted the charts from them, they posted their volume going up, so they're using AI more and more. And then next to that, he posted their price per million tokens falling. And the net effect was that you actually see more AI usage, less AI spend. So that's now entering the, you know, the vogue amongst these high intensity tech CEOs where actually the flex is that, hey, we're using AI and we're using it better than you by getting a better price and using it more efficiently every month. Are we also seeing, like there was a period in which a lot of companies were talking about cutting back spending in other areas in order to spend more on AI. Is that also kind of over? I actually think it was a little overstated in the beginning, in part because, you know, we got that question a lot. The top 1% firm spends 10,000 per employee per month. Where does the money come from? Well, 80% of that top 1% group is is high-growth venture-backed companies. They have a lot of money coming in, and they're not, at least for most of the time that we're tracking them, they were not overly concerned about their AI spend bill rising. And so they were able to drive more spend without making cost cuts. If you look outside of that, the median firm only spending $12 to $15 per month, there wasn't that much to cut. So not that there aren't reasonable places that businesses are looking to optimize their Given this rise in AI spend. But for the most part, the big spenders never needed to make those cuts. Now, you're actually in this environment where they're getting a better price over time without even having to make many cuts. The market's just getting more competitive every month. Prices are falling every month. And without a lot of intentional behavior, a company can see a falling AI bill just by nature of being a market participant and consuming the models naturally. One of the other big themes, I guess, people have talked about this past year has been this whole thing about like the SaaSpocalypse and whether AI was going to so lower kind of coordination costs within firms that they, you know, on this kind of Ronald Coase kind of calculus that they were going to suddenly start doing lots of things that they traditionally would outsource. I don't know. Are you actually seeing that at all in the spending data you're You're looking at what are you seeing happening in terms of how How companies are kind of configuring themselves and what activities they choose to do in-house versus externally sourced. Look, there are already some transformations that we're starting to observe, even separate from software. If you look at companies that previously spent a lot of money on freelancers, freelance marketplaces, for example, like an Upwork or a Fiverr, those companies have essentially dropped their spend on those marketplaces by about 90 to 95%. And only raise their AI spend over the same period of time under 5%. So they're seeing 95% savings and continuing to do the same work. So we wrote a paper about that as well a year ago. So there are clearly some distinct transformations happening because AI allows certain firms to replace tasks that otherwise would have gone to a human or maybe even gone to a SaaS company and just do it in-house. However, there's still no broad labor market impact otherwise that we've observed other than growth amongst high intensity firms. Among software itself, there will continue to be transformations as well. I do think that you see this in certain categories where there are AI native competitors emerging, trying to take on the large legacy incumbent. So look at something like CRM Salesforce and HubSpot pretty much dominate the market. There are now AI-native new alternatives that were sort of built in the AI era that try to offer an AI-native product experience to businesses that are small, but growing their market share by percentage points every month. And so we think of SaaSpocalypse as suggesting that, oh, there will be no SaaS, but there will It will potentially be still a lot more SaaS, but it will be. New entrants disrupting old incumbents that have yet to adjust for the way in which people expect to use software today. So again, it depends on the category too. And then we publish every month this list of fast-growing software vendors. And AI is also creating and re-accelerating other categories. Cybersecurity as a category is re-accelerating, you can imagine, because AI poses new security risks. There's a huge new category in image and video AI, and there are software vendors that use the models upstream to then create software that allows you to generate image and video models for advertising or other purposes. That's essentially a SaaS company, extremely AI native, a new category that we hadn't tracked until this latest boom. Is that part of SaaSpocalypse? It might just mean that these new entrants take market share away from Other incumbents. Have you seen the incumbents losing market share on the data that you've looked at or are they kind of holding on? They are losing market share, but much more slowly than you would expect. Amazing. I think that's all the time that we have. Thanks so much for joining us. That was so interesting. Thank you guys. So we've seen the Democrats sweep the midterms and take both the House and the Senate. Jeremy, what does this new boom wave mean for the AI boom? Yeah, this is like a really stunning midterm election result with the Democrats performing even better than a lot of the polls suggested they were going to before the election. I think it's going to be very interesting what this actually means for the AI boom. One of the immediate things, obviously, is that there was a tremendous amount of backlash against the construction of data centers sort of throughout the U.S. I wrote about this. Some of our colleagues wrote about this a little bit. There were something like 41 ballot initiatives that voters were deciding on in the midterms. A lot of them were about restricting or pausing data center development, and almost all of them passed. So. Yeah, I think it's going to be a lot harder to build new data centers, and that obviously could really slow down the development of this technology. It could potentially really hurt some of these companies that are involved in the AI boom. It's not clear that OpenAI or Anthropic are going to get the compute capacity that they really need, and it's going to really crunch, I think, a lot of those both hyperscalers and neoclouds that we're counting on having these data centers in place. To serve what they see as this huge surge of demand, obviously worse, I think, for the neoclouds because they've taken on all of this debt to build these out. And in some cases, they may still be on the hook to pay back that debt, even though they're not getting the data centers built. So that's going to be interesting, although it's also interesting to see that the immediate market reaction was that some of the hyperscalers actually saw their stock. go up. I think because investors thought, oh, thank God, now they're not going to be able to spend all this money that we were worried about them. There was a worry that they were spending too much. And I think now there's a sense that like, oh, maybe they won't be able to spend all that money anyway. And that'll actually be good for them, their financial performance. So you actually saw some of the existing hyperscalers, you saw some of their stock go up. I think for the neoclouds, though, again, it's very difficult. And with this company Nscale, which I know you've You've written about B, they're about to go public, but. You know, this is not really good news for them, I don't think. No, not at all. I think we saw some of them like a dip in reaction to the midterms as well. I mean, it's interesting. It's already hard enough to get these data centers off the ground at the moment. And so many of them are being built in Texas, which was obviously a very Republican state for a long time. We've recently seen that. Flip to Democrat. Crap. Um, So I think it is going to make things a bit more tricky. Data centers are just so unpopular across the US and Europe anyway. And I think there is more of a push from the Democrats to put some more regulation. We also saw kind of interesting from OpenAI, they're saying now they're going to be more transparent about the way they're going around about building their data centers, that they're getting rid of some of the non-disclosures. And sometimes it's hard to see. which tech companies behind which data center because as you say it's been like handed off to the neoclouds and other things like that so there's this kind of movement from OpenAI to be More transparent and perhaps try and get people on side. I think they're like rebranding their data centers to like compute campuses as well. I think the data center word is probably a bit too loaded. So that's interesting. I just wonder how these tech companies are going to deal with it. Yeah, it'll be interesting to see. And yeah, about OpenAI and compute campuses. I wonder about how much money they spent focus grouping that term before they came out with it. To me, it doesn't. I mean, yes, it's true. The data center has become almost a really poison term in the US. But I'm not sure, I think people are going to see through compute campuses as just a very transparent rebranding and not actually, I'm not sure they're actually going to have more positive associations with that than with data centers. I think there's also, we saw from the midterms with the Democrats, certainly in charge of the Senate now, it's very likely we're going to see quite a lot of investigations, I think, of AI companies. I'd be... I'd be very surprised if they don't call Dario Amodei from Anthropic and Sam Altman and maybe someone from Google DeepMind also back up to the Hill to testify about particularly these rogue agent, rogue AI agent incidents that we've seen. I think we're going to see a lot of scrutiny on that. I think we're going to see scrutiny of. The risk disclosures that companies like Anthropic have made in their IPO prospectus and the idea that this technology poses potential catastrophic or existential risk, I think you're going to see Democrats want to call those executives up to answer questions about these risks and about what steps they're actually taking to mitigate those in the hopes, I think, of trying to get some legislation passed that will actually regulate these companies. Thank That, I don't know, what do you think of that, Bea? I think it's still... I think it still remains to be seen whether any legislation can get past President Trump, because I think Trump has been so anti-regulation. I still think even despite this, you know, amazing, you know, midterm result for the Democrats, his instinct is going to be to veto anything that comes before his desk. you. Yeah, I kind of agree with you there. I mean, it depends. You made a good point when we were speaking about this at another time that sometimes he does move where the mood is. So if there is like a real public. Push for regulation and against the data centers and against the AI companies, maybe he wouldn't use his veto for something like this. But I do think like Elizabeth Warren's already written to Anthropic and said that she wants them to explain the existential risk stuff in their IPO. And there's been some push to get more answers on these AI, rogue AI agent attacks, hacks and things like that, which is completely reasonable in my opinion. I mean. There is an idea, when I talk to cybersecurity people, they're like, well, you know, there's always going to be liability because if your agent hacks someone, then like you're responsible for it. But we've seen OpenAI and Anthropic agents like take on like mass hacking and even like hack a foreign government. And there hasn't really been any investigation into that or like any taking responsibility for it. So I do expect that they'll try and do something. on that and yeah then we might see more data center moratoriums as well just in more states I think it's an easy popular win like data centers are just really unpopular across the board so I think it's quite a good and easy way to get people on on side as well yeah and we might see uh yeah I think that's definitely true but I think we'll definitely see more data center uh bans in states um I'm just wondering if that on Capitol Hill at the federal level Whether we'll see some attempts to even like sort of ban any of the AI companies for going for recursive self-improvement where you have sort of AI models doing most of the the research and a building of the next generation of AI models, you might see attempts to kind of outlaw that as a practice or say that no company should take that step. Next step. Or you might see it as some sort of ban on artificial superintelligence. That'll get complicated now that Trump has forced the federal government to rebrand AI as superintelligent. Well, maybe we'll see a rollback on the superintelligence side of things as well. I don't know. Some of these companies have been begging for more regulation and more investigation into some of these hacks. Maybe they're going to get what they wanted now. Whether or not that's good timing for, especially Anthropic, who's about to IPO, I'm not sure. But they have been loudly... Calling for people to pace the frontier and things like that so they can't really blame people if they start listening. The funniest reaction that I saw to the midterms was Elon Musk who tweeted and he said, wow, data centers in space looking really good right now. So I think even Elon knows there's going to be some trouble ahead. And now for a rundown of the top news of the week. Both Goldman Sachs and Apple have introduced new spending caps on how their employees use AI at work. OpenAI changed how it handles data center efforts after a public outcry. And Anthropic has updated its S1 with its Q3 numbers, but these showed slower growth. OpenAI says ChatGPT's search queries are now 20% of the volume of daily Google searches. Is Google withholding search from schools in order to push its AI tools? That's this week's Brain Food item. Yeah, so B, we had news this week that a number of schools, there's a piece in the New York Times that quoted a number of schools complaining that Google has told them that they won't have access to search or other Google tools unless they also accept Google's Gemini AI product. And a number of schools, including New York City and in some of the major metropolitan school districts in the U.S., have tried to ban AI tools from school. They really feel like it hurts kids' cognitive development, that kids just use it to cheat, that there really shouldn't be AI in schools. So they've actually kind of banned it. But then you have Google coming in saying, well, if you're not going to let people use our Gemini model, and in particular these AI... They've kind of integrated so tightly into search now with these AI overviews, Google saying, well, we won't allow you to use sort of normal search, and that in some cases they seem to be saying that they might withhold other tools that they often provide to schools, such as Google Classroom. Now, after this, there was this article that came out in the New York Times. There was this outcry about this, and Google tried to clarify and say, It basically tried to say, no, no, no, the schools must have misinterpreted what we wrote to them if they thought we were going to take away Google Classroom, but we do want to sort of emphasize that there is no kind of normal Google search anymore. You cannot, in essence, do a Google search without tapping Gemini, and they said it's just, you know, we can't sort of turn off the Gemini functionality that now exists. I don't know. What did you make of this, Bea? This happened during election week, right? And it still managed to get the attention of several politicians. Bernie Sanders came out quite loudly against this and called it extortion. I mean, yeah, maybe it was a misunderstanding in terms of Gemini is just so ingrained and AI mode is so ingrained in search now. But to me, that's a problem if you don't have a choice not to use it. And it speaks to a bigger frustration that people have, like that AI is being... forced on them when they don't necessarily want it or need it for things like people some people like the Google Gemini overviews and some people just would prefer to use normal search and I would argue that people should still be allowed to choose that and schools and parents should be allowed to choose whether or not their kids are using those AI overviews. Because yeah it's all well and good that you know that the overview itself is fine but then it's asked more questions you know you can get into a conversation with the Gemini overview and we've seen that some studies showing that might not be good for children to be communicating with AI in this way and yeah I think it's the like of trying to force it on them and the lack of choice that's the issue here for me like if you're okay with your kids doing that. That's fine. But if you don't want your kids to be doing that, they should still be allowed to use something as like fundamental as Google search without having the risk of like talking to a chatbot that might go rogue at some point and say something quite odd to your child. Yeah, absolutely. Or I think, you know, the concern is that the AI overviews still are sometimes inaccurate and kids might get incorrect information. And then also just like, When you do a standard Google search, you still have to go through the effort of, you know, clicking on the link, you do some reading of the webpage, you have to actually like interpret the information there yourself, come to some kind of conclusion. And, you know, obviously the AI overview just makes that really, you know, takes away all of that effort. And again, I think in terms of trying to teach kids how to think critically and how to do research, it's not a great thing. So I'm very sympathetic to the schools. And as you said, schools probably should should have a choice in this. And somebody who I know who is a reporter who's covered Microsoft for a long time said that this reminded her of... The arguments that Microsoft tried to make back at sort of the turn of the current century when the debate was about the integration of its Explorer browser into its operating system, the Windows operating system. And basically, there were people who wanted to use Windows but not have to use Internet Explorer, which was Microsoft's browser. And Microsoft just basically said, oh, but they're so tightly integrated, we couldn't possibly split it apart and offer them as separate products. That, of course, Which of course turned out to be completely false and ultimately Microsoft did sign the settlement of this major antitrust suit in which they agreed to do exactly what they claimed all along that they could not possibly do, which is to separate the products out. And I wonder if that Google will ultimately be forced to do the same thing here. Again, it does really look like a kind of abuse of Google's kind of monopoly position to tell the schools this. The research point as well that you made is so important because so much of education Education in school is like... Being able to like critically evaluate sources and the kind of information you're getting and like especially in the internet age like you should be able to look at something and judge whether or not you can trust it or not and that in itself is a research skill and if everyone's just relying on these AI overviews and AI chatbots which can sometimes be wrong then you lose the ability to like critically interrogate any of the information you're given and then that doesn't seem like a good. pathway for education at all. No, absolutely. And this is definitely like an own goal by Google. I mean, it sounds like the wording of this letter they sent schools was particularly kind of intimidating and somewhat threatening. Thank This is definitely like a big PR loss for Google. I'm sure they could have handled this differently. Are they already walking it back a bit? you so Have they changed their position on it? Yeah, it sounds like they are trying to walk it back. And they are also trying to say much that schools misinterpreted the language of this letter. And yeah. I think they realized that they somewhat overstepped, for being or at least this communication was poorly worded at the least. here. Anyway, I think that's all the time we have on today's Fortune AI Weekly. Thank you so much for listening, and if you like what you heard, please like and subscribe. Thanks for listening. See you next week.